Currently only operating systems running on Linux kernels with the option CONFIG_PACKET_MMAP enabled. This feature can be found even back to the days of 2.4 kernels. Most operating systems ship pre-compiled kernels that have this config option enabled and even the latest kernel versions got rid of this option and have this functionality already built-in. However, we recommend a kernel >= 2.6.31, because the TX_RING is officially integrated since then. In any case, if you have the possibility, consider getting the latest kernel from Linus' Git repository, tweak and compile it, and run this one! A note for distribution package maintainers can be found at the end of the file. What additional tools next to your build-chain are required? - pkg-config: all - flex, bison: bpfc, trafgen - ccache (optional) - gpg (optional) - sparse (optional) What libraries are required? - libncurses: ifpps, flowtop - libnacl or libsodium: curvetun - libnetfilter-conntrack: flowtop - libnl3: netsniff-ng, trafgen - libpcap: mausezahn, netsniff-ng (tcpdump-like filters) - liburcu: flowtop - libcli: mausezahn - libnet: mausezahn The following libraries can optionally be used to provide additional functionality for certain tools: - libGeoIP >=1.4.8: astraceroute, flowtop, netsniff-ng - libz: astraceroute, flowtop, netsniff-ng What additional tools are recommended, but not mandatory after the build? - cpp: trafgen - ntpd: curvetun - gnuplot: ifpps - setcap: all It is common that these libraries are shipped as distribution packages for an easy installation. We try to keep this as minimal as possible. One-liner installation for *all* dependencies on Debian: $ sudo apt-get install ccache flex bison libnl-3-dev \ libnl-genl-3-dev libnl-route-3-dev libgeoip-dev \ libnetfilter-conntrack-dev libncurses5-dev liburcu-dev \ libnacl-dev libpcap-dev zlib1g-dev libcli-dev libnet1-dev One-liner installation for *all* dependencies on Fedora: $ sudo yum install ccache flex bison ccache libnl3-devel \ GeoIP-devel libnetfilter_conntrack-devel ncurses-devel \ userspace-rcu-devel nacl-devel libpcap-devel zlib-devel \ libcli-devel libnet-devel After downloading the netsniff-ng toolkit, you should change to the repository root directory: $ cd netsniff-ng/ The installation (deinstallation) process is fairly simple: $ ./configure $ make # make install (# make uninstall) In order to remove all build files from the source tree: $ make clean In any case "make help" will give you some pointers of what can be done. To bring the source tree into a pristine state, there are two options. The first one will remove all build and build config file, the latter will also remove any manually added files: $ make distclean ($ make mrproper) You can also build/install/uninstall only a particular tool, e.g.: $ make trafgen # make trafgen_install (# make trafgen_uninstall) If you want to build all tools, but {curvetun,mausezahn} (i.e. because you don't need the tunneling software and the NaCl build process lasts quite long): $ make allbutcurvetun (allbutmausezahn) # make install_allbutcurvetun (install_allbutmausezahn) (# make uninstall) In order to build curvetun, libnacl must be built first. A helper script called nacl_build.sh is there to facilitate this process. If you want to build NaCl in the directory ~/nacl, the script should be called this way: $ cd curvetun $ ./nacl_build.sh ~/nacl There's also an abbreviation for this by simply typing: $ make nacl This gives an initial output such as "Building NaCl for arch amd64 on host fuuubar (grab a coffee, this takes a while) ...". If the automatically detected architecture (such as amd64) is not the one you intend to compile for, then edit the (cc="gcc") variable within the nacl_build.sh script to your cross compiler. Yes, we know, the build system of NaCl is a bit of a pain, so you might check for a pre-built package from your distribution in case you are not cross compiling. If NaCl already has been built on the target, it is quicker to use nacl_path.sh this way: $ cd curvetun $ ./nacl_path.sh ~/nacl/build/include/x86 ~/nacl/build/lib/x86 Instead of libnacl you can also use libsodium which itself is a portable, cross-compilable, API compatible drop-in replacement for libnacl. In order to use libsodium you need to provide some little help when running the configure script, so it will be able to detect the needed header files. Additionally you need to configure the build to not link against libnacl but against libsodium. In the case libsodium was installed into /usr/local you run the netsniff-ng configure script this way: $ NACL_INC_DIR=/usr/local/include/sodium NACL_LIB=sodium ./configure Alternatively you can use pkg-config for determining the include dir by setting the NACL_INC_DIR variable for the configure script like this: $ NACL_INC_DIR=$(pkg-config --variable=includedir libsodium )/sodium When done, netsniff-ng's build infrastructure will read those evironment variables in order to get the needed paths to NaCl. If you're unsure with any make targets, check out: make help In order to run the toolkit as a normal user, set the following privilege separation after the build/installation: $ sudo setcap cap_net_raw,cap_ipc_lock,cap_sys_admin,cap_net_admin=eip {toolname} For cross-compiling netsniff-ng, the process is fairly simple. Assuming you want to build netsniff-ng for the Microblaze architecture, update the PATH variable first, e.g.: $ export PATH=/microblazeel-unknown-linux-gnu/bin:$PATH And then, build the toolkit like this: $ CROSS_COMPILE=microblazeel-unknown-linux-gnu- \ SYSROOT= ./configure $ make SYSROOT should be set to the path where your cross-compiled libraries are stored, i.e. the path you specify as prefix when configuring the library using `./configure --prefix='. Note that some adaptations might be necessary regarding the CFLAGS, since not all might be supported by a different architecture. Probably the most simple way would be to run make CFLAGS="-O2 -Wall". For power users we have a set of zsh auto completion files, have a look at all file with ending *.zsh. There's also a BPF vim syntax highlighting file in the tree called bpf.vim. For doing a debug build of the toolkit with less optimizations and non-stripped symbols, do: $ make DEBUG=1 For debugging the build system, actual commands are verbosly shown if every make target is executed with: $ make Q= A hardening option is also available via HARDENING=1 if needed. You can then build and install the toolkit into prefixed path like: $ make PREFIX= $ make PREFIX= install Additionally to setting the PREFIX you can set DESTDIR if you want to install the toolkit outside of your current root filesystem but into an alternative path. This is particularly useful when cross-compiling because most likely you want to install into a dedicated sandbox or into a mounted root filesystem for the target architecture and not into the build hosts root filesystem. $ make PREFIX= DESTDIR= install Thanks for maintaining netsniff-ng in your distribution. Further questions will be answered on the public mailing list. Last but not least, there is one small utility for advanced users that we have not integrated into the main build process. This is a minimal BPF JIT image disassembler for the Linux kernel. You can also find this tool in the Linux kernel Git tree under 'tools/net/bpf_jit_disasm.c' or within the netsniff-ng Git tree simply under 'bpf_jit_disasm.c'. To build it, execute: $ gcc -Wall -O2 bpf_jit_disasm.c -o bpf_jit_disasm -lopcodes -lbfd -ldl The rest is described in the file header comment itself, i.e. how to get to the BPF JIT code. /?id=cff2f741b8ee8a70b208830e330de053efd4fc45'>Diffstat (limited to 'drivers')